Mind and Artificial Intelligence
Course Seminar CS 344
Aditya Somani
Prashant Pawar
Sanyam Goyal
Shashank
IntroductionIntroductionAn approach to simulate mind
from AILimitations in the path
Step by step…Step by step…Symbolic SystemNeural NetworksNeurons vs. MicrotubulesQuantum Physics
Symbolic SystemSymbolic SystemThe philosophy behind is that human
intelligence is rational, and can be represented by logical systems incorporating truth maintenance.
Formal system consisting of symbols.
Used patterns and rules.Knowledge is represented in formal,
symbolic form.Eg . TheoremProver
Symbolic SystemSymbolic SystemLearning was lacking in symbol
system.Model, based on the neuron-
network in the brain.Neural-networks
Neural NetworksNeural Networks
• Model biological neural systems.• Philosophy: Evolution and logical systems. Whatever works, works! Irrationality of mind. • Make ever changing decisions about what rules to follow.
Learning in BrainLearning in Brain
• Message passing. • If the total input of neurotransmitters to a neuron from other neuron exceeds some threshold, it fires an action potential.
Synaptic terminals
Courtesy ::www.wikipedia.org
Learning in BrainLearning in Brain
• Synapses change size and strength with experience.• When two connected neurons are firing at the same time, the strength of the synapse between them increases.
Modeling a NeuronModeling a Neuron
• Can be modeled as a graph where cells are nodes and synaptic connections are represented as weighted edges between the nodes.• Model net input to the jth cell as
where oi is the output of each
neuron connected to j.
1
32 54 6
w12
w13w14
w15
w16ii
jij ownet
Courtesy ::www.wikipedia.org
Modeling a NeuronModeling a Neuron
• oi is given by
where Tj is threshold for neuron j.
ji
jjj Tnet
Tneto
if 1
if 0
Neural ComputationNeural Computation
• Network is organized in layers made of nodes.• Training examples are given in the form of an output given a set of known input activations. • Recognize cat by examples of cats.
Courtesy ::www.wikipedia.org
Learning in Learning in Backpropagational Neural Backpropagational Neural NetworksNetworks
• Supervised process with cycles of input examples. • Occurs with forward activation flow of output and backward error propagation. • Gradient descent along the steepest vector of the error surface towards a global minimum of error. • Speed and momentum.
Neural ComputationNeural Computation
• Can be used to compute logical functions.• Can simulate logical gates: AND: Let all wji be Tj /n, where n is the number of inputs. OR: Let all wji be Tj
NOT: Let threshold be 0, single input with a negative weight.• Can build any circuit and machines with such circuits.
Strengths & WeaknessesStrengths & Weaknesses
• Massive parallelism will allow computation efficiency.• Behavior emerges from large number of simple units.• Flexible long-term memory.• Captures a variety of relations overcoming assumptions of linearity, independence etc.
• Require an adequate training dataset.• Training can be quite slow.• High error rate. • Black box.
Neurons vs. MicrotubulesNeurons vs. MicrotubulesNew models for consciousness proposed
in brain.Can we achieve self aware computers
(Singularity )with neurons ?
Neurons vs. MicrotubulesNeurons vs. MicrotubulesThe belief behind adopting neural
networks was that all the important action in the brain takes place using neurons .
But what about consciousness , is It handled by neurons ??
Studies of ParameciumStudies of Paramecium• A number of studies have observed
Paramecium swimming and escaping from capillary tubes in which they could turn around.
• They take less and less time as we keep repeating the experiment
Studies of ParameciumStudies of Parameciumit is hard to explain how a one-celled
animal like paramecium with “NO neurons” can learn if we say that neurons are responsible for learning in multi-celled animals
The theory to explain this is that the nervous system of the Paramecium (cytoskeleton ) is responsible for doing all this computing .
CytoskeletonCytoskeletonA collection of hollow fibers called
microtubules made out of a protein called tubulin.
The microtubules consist of molecules of tubulin that can be in two different states depending on the presence or absence of an electron, a nice digital system.
Is Singularity AchievableIs Singularity Achievablethree reasons to say why singularity is not near :- 1)The mind is synchronized (But how??)
(i) how these ever-shifting, widely distributed groups of neurons in sync? Not answered yet!
this leads to doubts in taking neural-network 2)The brain is faster (so what ??)
In neural network, AI assumes that the neuron is analogous to a single computer bit. But later it was found that each neuron is supported by a additional circuitry., Which AI do not take care.
3) Anesthesia (contradicts the assumed fact that consciousness arises from firing neurons)
Microtubules to Quantum Microtubules to Quantum ComputingComputingPenrose is among a number of
researchers proposing that – ”there is quantum computing going on in the brain and quantum effects are responsible for the flash of insight phenomenon.”
Penrose proposes that quantum computing is happening in the microtubules of neurons , which is responsible for consciousness
Mind and Quantum Mind and Quantum PhysicsPhysics
Penrose and Gödel's Penrose and Gödel's Theorem:Theorem: Gödel's Incompleteness Theorem: with any set of
mathematical axioms, it is possible to produce a statement that is obviously true, but could not be proved by means of the axiom.
Penrose's Argument(The Emperor’s New Mind ,1989): The theorem showed that the brain had the ability to
go beyond what could be achieved by axioms or formal systems
Mind had some additional function that was not based on algorithms
But, a computer is driven solely by algorithms Brain could perform a function that no computer
could perform Called idea of non-computable functioning
Penrose: Brain and Quantum Penrose: Brain and Quantum PhysicsPhysicsNot all human intelligence is
algorithmicPhysical laws are described by
algorithmNot easy to come up with physical
properties or processes that are not described by them
How do then we explain the implied superiority of human brain?
Quantum Physics!
Quantum Theory: Coherence Quantum Theory: Coherence and De-coherenceand De-coherence Sufficiently isolated quanta : can be viewed as waves;
waves of probability(position, momentum). Quanta subject to measurements, interaction with the
environment, wave characteristic lost, and a particle is found at a specific point.(position waves).
Called collapse of the wave function No cause-and-effect process No system of algorithms can describe the choice (of
position)for the particle. Seems to suit the search But randomness Not a promising basis for mathematical
understanding.
Objective Reduction: The Objective Reduction: The IdeaIdea Penrose's proposition of a new form of wave function
collapse. Relativity: mass causes curvature in space-time fabric Space time fabric, continuous on relativistic scales but a
network on quantum scale Reconciliation of relativity and quantum physics Proposition each quantum superposition has it’s own
curvature Blisters on the spacetime fabric ~( 10 -35 meters, Planck
scale) Above Planck scale gravity comes into effect, system
becomes unstable Collapse so as to choose just one of the possible values Called Objective Reduction
Objective Reduction: The Objective Reduction: The Time FactorTime FactorEt = h/2pi; E = gravitational self-
energy , t = time to collapseThe greater the superposition the
faster is the ORFor electron 10 million years, for
a kilogram object (10-37 seconds)For usual objects order relevant
to neural processing time.
Objective reduction: the Objective reduction: the scopescopeChoice of states neither random,
as are choices following measurement or de-coherence, nor completely algorithmically.
Orch OR model: Bringing Orch OR model: Bringing Quantum Physics to BrainQuantum Physics to Brain Do we do Quantum Computing? Microtubules may be supporting quantum processing: Shadows of the
Mind (1994), Penrose/ Hameroff comprised of subunits of the protein, tubulins: contain hydrophobic (water
repellent) pockets hydrophobic pockets from different tubulins within two nanometers of one
another close enough for the π electrons of the tubulins to become Quantum
Entangled Quantum Entanglement:
◦ “a state in which quantum particles can alter one another‘s properties instantaneously and at a distance, in a way which would not be possible, if they were large scale objects obeying the laws of classical as opposed to quantum physics”
◦ principle of non-locality ◦ the EPR experiment
Hameroff's proposition: large numbers of the π electrons can become involved in a Bose-Einstein condensate
Bose Einstein Condensate: These occur when large numbers of quantum particles become locked in phase and exist as a single quantum object
happens usually at a very tiny scale but can be boosted to be a large scale influence in the brain
Orch OR Model: making it Orch OR Model: making it bigbig Gap junction:
◦ intercellular connection between cells
◦ allows various molecules and ions to pass freely between cells
◦ in addition to the synaptic connections
proposition: condensates in microtubules in one neuron can link with other neurons via gap junctions, using quantum tunneling
allows the Bose-Einstein Condensates to cross into other neurons
extend across a large area of the brain as a single quantum object
when condensates in the brain undergo an objective reduction of their wave function, there is an instance of consciousness
brain gets access to a “non-computational process embedded in the fundamental level of space time geometry”
The AHA moment!
Orch OR Model: EpilogueOrch OR Model: Epilogueproposition: Orch OR causes gamma
synchronization microtubules both influence and are
influenced by the conventional activity at the synapses between neurons : Orchestrated OR
Orch OR Model: Criticism and Orch OR Model: Criticism and Counter-Criticism Counter-Criticism Penrose's hypotheses: yet to be supported by
experimental evidence Tegmark: microtubule quantum states would persist for
only 10-34 seconds at brain temperatures far too brief to be relevant to neural processing, rapid
decoherence Hameroff Retaliates:
◦ Tegmark’s model incorrect: 24 nanometers is too far
◦ Shielding by water molecules
◦ pumped into a coherent state by biochemical energy
◦ quantum error correction
"Some people see that Penrose is obviously right. Some people see that Penrose is obviously wrong. What's obvious then is that the issue is not obvious" -- Donald R. Tveter
Consciousness and QPConsciousness and QP
Earliest propositions: James Jeans(physicist), Alfred Lotka(biologist), 1920's
Two major schools of thought:◦ Copenhagen Interpretation (Penrose et al.)◦ Bohemian Interpretation (Bohm and party)
Copenhagen Interpretation:◦ The wave function is : "complete and literal description of
the state of a quantum system“◦ Reality exits only when you measure it.◦ Schrödinger's 'cat experiment‘◦ Possible explanations:
consciousness collapses the wave function and thereby creates reality
whole universe must have existed originally as "potentia" in some transcendental realm of quantum probabilities until self conscious beings evolved
Consciousness and QPConsciousness and QP
Bohemian Interpretation:◦real existence of particles and field◦ Implicate order: a vast ocean of
energy on which the physical, or explicate, world is just a ripple already present in quantum physics: the
quantum vacuum or zero-point field perhaps something like the Addvait principle
in the Indian Philosophy
-
ConclusionConclusion - Mind offers a "model model" to pursue the
goal for human-like intelligence. - However, the exact working of human
mind is far from trivial.- Continuous research efforts should help us
get closer and closer to the knowledge of the actual principles of the human brain
- We have already covered a long distance: Symbol Systems -Quantum Physics
- long way to go!
ReferencesReferences1. http://www.wired.com/medtech/drugs/magazine/16-04/f
f_kurzweil_sb2. Donald R. Tveter
http://www.dontveter.com/caipfaq/systems.html3. Consciousness, Causality, and Quantum Physics: David
Pratt, Journal of Scientific Exploration, 19984. http://www.en.wikipedia.org/wiki/Orch-OR5. Orchestrated Objective Reduction of Quantum
Coherence in Brain Microtubules: The "Orch OR" Model for Consciousness ,Robert Penrose and Stuart Hameroff 1996
6. http://www.cis.temple.edu/~vasilis/Courses/CS44/Handouts/neural.html and various other online resources.
Thanks!Thanks!
Questions?Questions?